Outline for a causal model of traffic conflicts and crashes

Abstract

Road crashes tend to be infrequent but with nontrivial consequences, leading to a long-running interest in identifying surrogate events, such as traffic conflicts, that can support a timely recognition and correction of safety deficiencies. Although a variety of possible surrogates have been suggested, questions remain regarding how crashes and surrogates are related. Using recent developments in causal analysis we propose a simple model which represents crashes and conflicts as resulting from interactions between initiating conditions and evasive actions, and then use this model to identify relationships between these types of events. Our first set of results expresses the probability of a crash as a mixture of probabilities over different sets of initiating conditions, where the mixing probabilities are governed by the evasive action. Our second set of results considers situations where sampling is restricted to non-crash events, and gives conditions where these truncated probabilities can serve as proxies for crash probabilities. We then illustrate how trajectory-based reconstruction can be used to classify initiating events with respect to severity, and to estimate a proxy for the crash probability from a set of incompletely observed non-crash events.